Multi-agent systems, LLM applications, generative AI, and computer vision — delivered to teams that need them in production.
I've consulted for early-stage startups and enterprises alike, delivering multi-agent platforms, RAG pipelines, and generative-AI systems that ship and stay shipped. The same hands writing that code are also publishing at ICLR, ICML, KDD, CVPR, ACL and Q1 journals — so recommendations are grounded in current research, not engineering hunches alone. In engagements that run from advisory retainers to fixed-scope builds, the deliverable is code, not slides.
Building multi-agent systems with LangGraph, LangChain, and LangSmith — intelligent chatbots, coding agents for Excel-based data processing, and retrieval-augmented generation applications deployed on Azure (blob storage + indexing). Driving code reviews, leading projects, and mentoring junior developers.
Designed graph-based agent workflows in LangGraph/LangChain to analyze email threads and track purchase orders. Long-term memory, planning, reasoning, reflection, and self-improving feedback loops; a critique agent for decision evaluation; human-in-the-loop validation; full evaluation framework in LangSmith.
Monetized, fine-tuned, and researched state-of-the-art LLMs and CV models across image/video captioning, semantic search, text-to-image generation, object detection, deblurring, super-resolution, and 3D scene reconstruction. Built scalable services in Flask/FastAPI/.NET with MongoDB, Redis, and SQL; shipped recommendation systems for sales optimization.
Designed infrared and optical-range imaging systems for space and defense applications. Ray tracing, Fourier optics, classical CV, and deep learning across the optical-pipeline / signal-processing stack.
Designed consumer display systems incorporating ray tracing, computer vision, and deep learning. Supporting tooling in Python and React/Node.
Production multi-agent workflows in LangGraph and LangChain — long-term memory, planning, reasoning, reflection, and self-improving feedback loops. Critique agents, human-in-the-loop validation, full eval in LangSmith.
Chatbots, coding agents, document-grounded retrieval. Vector stores (Qdrant, Weaviate, pgvector) and graph stores (Neo4j) where each fits. Azure / GCP / AWS stacks with blob storage, indexing, and the boring infrastructure that makes a demo into a deploy.
Text-to-image, super-resolution, deblurring, video captioning, 3D scene reconstruction. Fine-tuning, deployment, scaling — productionized for media platforms at scale.
Drawing on five years of computational-imaging work at Aselsan and Arçelik. Optical pipelines, real-time CV, anything where the camera and the model have to be designed together.